Institute of Applied Informatics Wroclaw University of Technology Wroclaw, Poland. E-mail: halina.kwasnicka@pwrwroc.pl
Evolutionary algorithms become very popular due to their searching skill in a solution space. The problem arises when we try to adjust used genetic operators and parameters. In literature one can find various, often sophisticated new genetic operators, specific for the particular task. In the proposed method a user can use a number of cooperating and specialized genetic algorithms with simple genetic operators and presumed parameters, but an intelligent agent takes care of tuning the parameters. The agent tunes parameters dynamically on the basis of observed results. We have defined a number of measures used by the agent as inputs for a fuzzy control system. The set of fuzzy rules can be defined using experts' knowledge.
The main advantage of the proposed system is releasing of GAs users fromonerous duty – the determination of genetic operators and values of their parameters. It is usually a time consuming task requiring extensive experience. The proposed system is flexible enough to solve the problems which potential solution can be represented as a string of real values.
The paper presents the initial studies as well as the final proposition: GAAgent, an ensemble of cooperating genetic algorithms controlled by a set of fuzzy rules. The exemplary results are presented and discussed.
